Official AMD Radeon RX 7000 Series (RDNA 3) Thread

[Exposed]

But 7900XTX is not CPU limited? How do we know?


The Digs were funny, yes they would have shown them losing in benchmarks. Does it matter if the card is priced at $999. Remember 4870?



OK we are on the same page finally. Im going to bed, but good times my friend.

Times like this I wish I could fast forward to December 13th (or whenever NDA expires) to find out the real deal ... but then I realize I'd be one month older for no reason

[Nagorak]

The price is a lot more palatable to me than the RTX 4090, or even 4080 at $1200. Even if it's a little bit slower and FSR is inferior to DLSS I will probably be in the market for a 7900 XTX.

Ray tracing still isn't that important to me-- 3090 Ti level would be sufficient. I still mostly need performance for VR where ray tracing is nonexistent and DLSS is close to nonexistent. At 1440P I will be CPU limited a lot of the time anyway, so even a bit shy of 4090 won't result in much difference. (Not interested in upgrading my monitor for a while yet--hoping to hold out either for mini-LED or OLED.)

All things being equal I'd rather have the performance to just run native without DLSS/FSR anyway, barring the rare game with a bad TAA implementation.

[pax]

Some perf estimates puts the 7900xtx as even with the 4090 in non rt games... but at a much lower tbp. Thats pretty impressive if true...

Hardware Canucks did that in their vid.

[yt]HFR0pyXtVrU[/yt]

It seems all the RT perf uplift was in the changes made to the RT cores because they maybe didnt add any per CU unlike the more than doubling of shaders per CUs. So 6950xt was 5120 shaders in 80 CU's vs 12000 shaders in 96 CUs in the 7900xt but seems they didnt double the RT units per CU.


Core clock increases in aibs or ocing the Ref might give 10-20% more RT perf if they can do it... Linear with raster perf.

[Nagorak]

I'm wondering if Nvidia inadvertently did AMD a favor by setting the board power of the 4090 so high.

It's looking to me like the 4090 would be slightly faster at even power, since it doesn't seem to drop that much performance if you cut the power limit by 100w. So at 350w vs 350w the 4090 might still be 5-10% faster. But instead Nvidia pushed it way into diminishing returns which only netted another ~5% or so performance. So, now AMD can release their cards at 350w and look good on both price and efficiency, instead of just on price.

The 4080 is also looking bad to me at $1200 being so cut down compared to the 4090. Based on Nvidia's slides it's going to be slower than the 7900 XTX (and possibly XT), at least in raster. Kind of a tough sell at that price, IMO. Guess we'll have to see how that shakes out. Good thing they "unlaunched" the 4080 12GB or that would have been downright embarrassing.

[koralis]

I mean, you can make light of it all you want, but it's pretty standard knowledge a 4090 is going to be CPU limited in many cases starting from 1440p and bleeding into 4k. So that will skew your "13 game average" results, and many sources already acknowledge this.

This tells me that these high end cards are *pointless* if they are cpu bound and the only way to get higher numbers is to fake it. Might as well buy a cheaper card... Which is probably a major factor in 4070 and 4060 being nowhere to be seen... They would have been benchmarked against the 4090 and the reviewers may have had a comment or 2 about the 4090 not being worth the price premium.

[Android1]

AMD Radeon RX 7900 Series GPUs "Designed to Scale Up to 3.0 GHz"

Lower than rumored.

The rumors about very high clocks on Navi 31 GPUs were wrong, but is there any chance we will see a high clock on custom RX 7900 graphics cards? According to PC World who were invited to the AMD RDNA3 launch event, the company said they designed Navi 31 GPU to scale up to 3.0 GHz, which is much higher than official clocks for RX 7900 series.

AMD RX 7900 XTX has a boost clock of only 2.5 GHz, while RX 7900 XT only goes up to 2.4 GHz. This is, of course, contrary to the rumors about 3.0 GHz clocks out of the box.

YouTuber ‘coreteks’ claims to have some information on the subject. With the preliminary drivers, board partners are only seeing 3% overclocking potential with custom Radeon RX 7900 series. This is indeed low, but it will probably be higher with the final drivers.

Unfortunately, no board partner except ASUS has unveiled their custom cards yet, and this brand does not even provide any clocks for any of their ‘factory-overclocked’ models:

[Screenshot of ASUS TUF Gaming Radeon RX 7900 XTX OC Edition 24GB GDDR6 specs in article. —Android1]

Sources: VideoCardz.com, ASUS, Coreteks (Twitter), PCWorld (YouTube), TechPowerUp

Click HERE to discuss this topic on our forums!

[Android1]

Thread title changes to reflect the Radeon RX 7000 Series’s official status.

[pax]

Yeah only 3% left on the table is hardly realistic. AIBs have been locked down by drivers because they leaked too much info and they had only thermal testing drivers to work with. Going to 450w will gain 10-20% for aibs. AMD isnt doing an nvidia thing where it leaves nothing for aibs to go for.

One tech tuber said the perf is only about 6.9% lower on average as it stands from the games amd used on its presentation. It maybe worse if other games are less performing in reviews.

[Android1]

AMD Radeon RX 7900 XTX Performance Claims Extrapolated, Compared to GeForce RTX 4090

AMD’s new flagship GPU will supposedly offer performance extremely close to that of the NVIDIA GeForce RTX 4090. Of course, this comparison is based on data that came from AMD, so I would take it with a large grail of salt.

The graphs put the Radeon RX 7900 XTX menacingly close to the GeForce RTX 4090. In Watch_Dogs Legion, the RTX 4090 is 6.4% faster than the RX 7900 XTX. Cyberpunk 2077 and Metro Exodus see the two cards evenly matched, with a delta under 1%. The RTX 4090 is 4.4% faster with Call of Duty: Modern Warfare II (2022). Accounting for the pinch of salt usually associated with launch-date first-party performance claims; the RX 7900 XTX would end up within 5-10% of the RTX 4090, but pricing changes everything. The RTX 4090 is a $1,599 (MSRP) card, whereas the RX 7900 XTX is $999. Assuming the upcoming RTX 4080 (16 GB) is around 10% slower than the RTX 4090; the main clash for this generation will be between the RTX 4080 and RX 7900 XTX. Even here, AMD gets ahead with pricing, as the RTX 4080 was announced with an MSRP of $1,199 (exactly 20% pricier than the RX 7900 XTX). With the FSR 3.0 Fluid Motion announcement, AMD also blunted NVIDIA's DLSS 3 Frame Generation performance advantage.

Source: TechPowerUp, @harukaze5719 (Twitter)

[bill dennison]

[yt]YSAismB8ju4[/yt]

[Trunks0]

It's looking like a price/performance champ. Which is a nice thing to see given the current environment... but it's all still expensive. Ah, probably the best we can hope for with the way things are right now.

There's nothing wrong with it IMO as long as latency is not affected for the worse.

Ive turned this on my old LG TV and enjoyed it even with the added latency it brought. Glad AMD is going this route too, so now we have a better native method rather than the TV without the drawbacks of traditional frame interpolation.

And there is that. It's really not a big deal, short of the typical forum banter, but that gets a whole lot less intense if everyone just has the feature. Now we just need Intel to do the same.

[Android1]

It's looking like a price/performance champ. Which is a nice thing to see given the current environment... but it's all still expensive. Ah, probably the best we can hope for with the way things are right now.

Driver quality will be key. Faster means nothing if you have constant problems.

As for everything being so expensive, I totally agree. Ugh.

[bill dennison]

AMD's Navi 31 Might Clock to 3 GHz, Partner Cards Will be Able to Overclock

https://www.techpowerup.com/300655/amds-navi-31-might-clock-to-3-ghz-partner-cards-will-be-able-to-overclock

[Trunks0]

Driver quality will be key. Faster means nothing if you have constant problems.

As for everything being so expensive, I totally agree. Ugh.

Yup and this might be one of the first times in a very long time where I personally jump at the start of a fresh series of cards and buy a flagship for myself. If nothing else it will be a learning experience to do it again.

[acroig]

Yup and this might be of the first times in a very long time where I personally jump at the start of a fresh series of cards and buy a flagship for myself. If nothing else it will be a learning experience to do it again.

About time!

[Android1]

Yup and this might be one of the first times in a very long time where I personally jump at the start of a fresh series of cards and buy a flagship for myself. If nothing else it will be a learning experience to do it again.

You might want to wait a bit. Let the paying beta testers like Alex suffer with the teething pains.

[AllexxisF1]

So I can play Return To Castle Wolfenstein at 500FPS on the 4090 or 492 FPS on the 7900XTX? - and the 7900XTX is $600 cheaper.

TOUGH DECISION.

[Trunks0]

You might want to wait a bit. Let the paying beta testers like Alex suffer with the teething pains.

normally I would. But its been such a long time since I've done it... so bring on the pain. See if it ends up making me switch back to what I usually do .

[pax]

https://www.techpowerup.com/300655/a...e-to-overclock

Apparently 2.8 is easy but 3 ghz is golden silicon. AIB's will certainly segment that for us. If you can get a 3 ghz part the 7900xtx might be faster raw raster than a 4090.

https://www.youtube.com/watch?v=XyHpTPLsgNA

[pax]

Explanation as to why amd is slower in RT:

https://twitter.com/AMDGPU_/status/1588599436363825152

AMDGPU
@AMDGPU_
Replying to
@Kepler_L2

Even at 4k 90fps 3 bounce
Only 2-8 billion ray-tri intersects per sec are needed which is well bellow RDNA2/3 ability

ray-box intersect is more relevant as youll need up to 250 billion which RDNA2/3 can do

Continued 1/2⬇️

AMDGPU
@AMDGPU_
·
3h
Replying to
@AMDGPU_
and
@Kepler_L2

The real issue is more complicated

Nvidia RT cores can traverse the BVH async avoiding thread vacancy

AMD RT cores cannot do traversal so its offloaded to shaders that operate in lockstep causing massive thread vacancy during traversal killing RT performance
2/2

So if its a shader function in some software layer maybe driver updates can address that in the future.

[CyanBlues]

luckily I took those days off for my bday, looks like I'll have to camp Microcenter lol

[Nunz]

luckily I took those days off for my bday, looks like I'll have to camp Microcenter lol

Lol, nice

[NIGELG]

Time to plan for an eventual return home to Team Red.

[acroig]

You might want to wait a bit. Let the paying beta testers like Alex suffer with the teething pains.

[acroig]

Time to plan for an eventual return home to Team Red.

[acroig]

luckily I took those days off for my bday, looks like I'll have to camp Microcenter lol

Lol, nice

Yes, rumors have it that day one inventory is less than 4090 was so a sell out is expected.

[Nunz]

We shall see. The mindshare for AMD is much lower. I don't know if they'll have similar demand, even against a $1600 card like the 4090.

[acroig]

We shall see. The mindshare for AMD is much lower. I don't know if they'll have similar demand, even against a $1600 card like the 4090.

Yeah with 4080 being dogpoo I'd say the 7900s will sell out.

[NIGELG]

It depends on benchmarks and prices of course. A 3070 is very capable at my resolution.

Let's see what both camps have for 699 bucks or so. I'm seeing 3080 Ti's for sale in my country, but I'm limited at 750 watts.

[NIGELG]

Yeah with 4080 being dogpoo I'd say the 7900s will sell out.

Price war for the lower tiers?

[the_sextein]

Here is a look at the new design. It's a leak so take it with a grain of salt.

https://videocardz.com/newz/alleged-amd-navi-31-gpu-block-diagram-leaks-out

Based on the data floating around regarding game performance the 7900XTX looks to be about 10% slower than a 4090 at rasterization. Which means it will probably trade blows with the 4080 in this area. RT performance appears to be a little behind Nvidia but it's more competitive this time than last time. FSR is also close but a little behind DLSS in Image quality. AMD is offering more VRAM but it's slower RAM and games will not need more than 16GB over the next 2 years. The card costs $200 less and AMD's FSR 3.0 with frame generation still has a little while before it hits the scene.

AMD's 7900XTX looks like it will be competitive with the 4080 but it still feels like AMD is dragging behind. They just trail blazed a new approach to GPU so that is exciting and hopefully as they tweak the new arc we will see it's advantages start to move in on Nvidia. They really are not that far off in any area but they are behind almost everywhere by a little vs the 4090 and they need to improve FSR and RT performance just a little more than Nvidia to catch up.

[Nunz]

I don't know why you'd think a 4080 will be 10% lower than the 4090 with ~30-40% less CUDA cores. I think the 4080 is going to be significantly behind the 4090.

NV is rectifying their mistake of making the 3080/3090 so close in CUDA core count, because that's what it was. Mistake.

[the_sextein]

Good point, I'm so used to Nvidia charging a premium for 10% more at the high end I forgot that they actually improved the 4090 over the 4080 by more this time. That would give AMD a nice rasterization advantage against the 4080 and while it may be behind in RT and FSR it is $200 cheaper as well. I'll wait for the official reviews to see how things pan out but that does make the 7900XTX look like a better deal than the 4080 in my book. It still doesn't solve everyone hate of the $1000 price point though. Personally I understand it as the cost of everything is going up and they are improving in more areas than just raster performance and it's still cheaper than Nvidia. I think the days of $700 to $800 high end are gone unfortunately.

[koralis]

I don't know why you'd think a 4080 will be 10% lower than the 4090 with ~30-40% less CUDA cores. I think the 4080 is going to be significantly behind the 4090.

NV is rectifying their mistake of making the 3080/3090 so close in CUDA core count, because that's what it was. Mistake.

If everything is cpu bottlenecked, those extra cores might not have a lot to do in real world usage. Pretty sure that's why the frame interpolation exists... Give the gpu something to do that doesn't involve the cpu at all.

[acroig]

It depends on benchmarks and prices of course. A 3070 is very capable at my resolution.

Let's see what both camps have for 699 bucks or so. I'm seeing 3080 Ti's for sale in my country, but I'm limited at 750 watts.

For a 3080Ti you need a good 850W PSU unfortunately. A good 750W will do fine for a 3080.

[acroig]

Price war for the lower tiers?

Undoubtedly! 4070 and 4060 will have a hard time competing with AMD's lineup.

[NIGELG]

For a 3080Ti you need a good 850W PSU unfortunately. A good 750W will do fine for a 3080.

Well maybe power consumption will be lower for the AMD equivalent of a 3080Ti or else it makes no sense for me to upgrade this generation.

Nvidia is power hungry.

[Nunz]

If everything is cpu bottlenecked, those extra cores might not have a lot to do in real world usage. Pretty sure that's why the frame interpolation exists... Give the gpu something to do that doesn't involve the cpu at all.

There are very few games that are CPU bottlenecked at 4K.

If that's the reason why it exists, that's a pretty poor reason.

[pax]

With the fire sales on the 6000's and the keynote that FSR3 may be enabled on them we might see them go past the 3080's in not just raster but RT in those games that use FSR3. That would be a remarkable change if that happens.

Would make the 6000's the one to reach for as the prices finally drop.

[Mangler]

The frame generation tech is probably going to be too "gen 1" in both camps in order to matter outside very specific cases.